HS 
ue | : 
  
  
  
  
  
Figure 6. Shaded model and orthomosaic of the Zea theatre. 
numerous archaeological sites have been mapped at one time or 
another, usually with geodetic surveys on behalf of the Ministry 
of Culture, which has produced an amassed wealth of available 
data, mostly planar. Depending, of course, on their accuracy and 
quality, one might well take advantage of such graphical data at 
an extended scale as exclusive control information to produce 
orthomosaics for innumerable sites, at least as basic archival do- 
cumentation. 
One could proceed by simply deriving planar control points XY 
from such 2D information and execute a ‘purely planimetric’ ad- 
justment. As, however, solution is impossible without some ele- 
vations, the procedure in a digital photogrammetric workstation 
could be as follows. First, a ‘relative’ phototriangulation is per- 
formed, i.e. the strip model is formed (in the system of the first 
image), providing for all control points their model coordinates 
Xyz, which can be correctly scaled. These elevations z, referring 
to a tilted model system of arbitrary z-origin, are then used with 
very small weight in control point triplets XYz for a new bundle 
adjustment. Outcome of such essentially planar adjustments will 
be the exterior orientation of the images, related to the arbitrary 
model z-origin. 
Experimentation has indicated that the resulting inaccuracies in 
elevation, due to small uncorrected model tilt, may be diminish- 
ed by using more realistic weighting. After the first solution, the 
O, à, K values for the reference image are used to determine a 
maximum value for the model displacement in z, at the control 
point remotest from the nadir N of the image. This value is used 
in a new bundle adjustment as better approximation for weight- 
ing model elevations. to provide new ®, À, K values. From these 
a final weight for each z is calculated as its expected uncertainty 
due to model tilt. The XY coordinates are first rotated by x, and 
their absolute AX and AY differences from N are then combined 
with the absolute ©, À values to give the error estimation 
s,= AX¢ + AY® 
used for weighting elevations in the final phototriangulation. In 
this sense, object relief is also taken into account. 
Both the described approach and ‘planimetric’ adjustment have 
been applied to the data of all four projects. Table 7 presents the 
RMS differences of the resulting control point coordinates from 
those estimated by bundle adjustment using full 3D control. Of 
course, all Z-values thus obtained have been first shifted to refer 
to the mean elevation of the control points. 
  
Table 7. RMS differences between full bundle adjustment and 
adjustments using only 2D ground information 
  
planimetric adjustment | with weighted elevations 
X(cm) Y(cm) Z(cm) | X(cm) | Y(cm) | Z(cm) 
Zea 0.5 1.1 16.1 0.3 0.5 7.0 
Ag. Marina | 1.3 1.2 23.4 0.6 0.6 4.6 
Aigosthena | 0.4 0.4 19 0.4 0.4 1.8 
Sparti 0.6 0.3 3.3 0.6 0.3 3.4 
  
  
  
  
  
  
  
  
  
It is seen that in the last two cases both approaches provide very 
good results. This is apparently due to the strong configurations, 
as the extension in depth was significant compared to the short 
imaging distances, and most object points were intersected with 
several rays. The tilt of the reference image is expected to play a 
role, too, and so does the form of the strip (which in 4g. Marina 
is long and narrow). In the first two cases, improvement is clear. 
Evidently — as also witnessed by the larger discrepancies in Z — 
small uncorrected tilts are still present in all cases. The distribu- 
tion of Z-differences, seen in Figs. 8 and 9, illustrate this effect. 
Even if elevations are ignored, however, planimetric orientation 
appears as sufficient for the projection on the horizontal plane. 
Indeed, no significant differences were detected between ortho- 
;41-